- Estimates correct MCS using EMA, kept between target thresholds
dl_bler_target_upper and dl_bler_target_lower (conf options) for 1st
retx and below dl_rd2_bler_threshold for 2nd retx
- Maximum MCS configurable
- on 3rd retx, reduce MCS by 5
1: there can be 2bits AckNack + SR in PUCCH F0
2: when allocating csi_pucch, zero out struct
3: Make unfree resources a warning
4: When asserting, explain the issue
- Configure PUCCH Format 0 to be in separate RBs (each UE has dedicated
resources)
- Configure RA preamble to be different per UE
- Configure CSI offset such that no two UEs have their report in the
same frame
Nb_ue is not unique among UEs: consider 2 UEs present. The last UE has
CSI idx "2". Now the first UE disconnects, and a new UE connects. It
will also get CSI idx "2", and I assume this is not what we want.
With the merge of fix_NR_DLUL_PF, gNB_dlsch_ulsch_scheduler() has no
nb_periods_per_frame variable. There are two changes here:
1) calculate this locally for initializing tdd_beam_association
2) do not pass it into schedule_nr_mib(). Further, use the following:
nb_periods_per_frame = slots_per_frame / slots_per_period
slots_per_frame / nb_periods_per_frame =
= slots_per_frame / (slots_per_frame / slots_per_period)
= slots_per_period
1) Only count new transmission as scheduled bytes (which is then
compared against the BSR)
2) When subtracting scheduled bytes after successful reception, subtract
TBsize from correct HARQ process
As the precedent commit, this commit deals with the realtime problems
that we currently have on the CI bench.
For DL HARQ feedback, nFAPI does not give us the HARQ process ID.
Instead, we have to figure this out "from timing", i.e., if we trigger
PUCCH reception, we will get a message with HARQ if we instructed the
PHY to do so. If we have real-time problems, it seems (I cannot verify:
don't have real time problems) that we do not get the nFAPI message with
HARQ feedback. Thus, we need to skip the HARQ processes that should have
been processed in the past, which happens in this commit.
At the time of this commit, the CI has realtime issues. This can mean
that the HARQ result is delivered in a delayed fashion. Before this
commit, we were asserting on whether the slot corresponds to the HARQ
process's expected feedback slot, but a particular feedback might be
delayed under realtime issues, and we actually don't care about the
timing since we get the HARQ process ID.
Instead, we loop through the HARQ processes for which we wait for a
feedback in the beginning until we have the right process (which did not
seem to cause a problem till now, but who knows.
This commit enables more slots so that the "real UE" preprocessor can
schedule them. The DL phytest preprocessor continues to only schedule
slots 1 & 3. This is because the OAI-UE needs to have PUCCH scheduled
at least six slots in advance. Since we do not support this (yet), but
assume that two slots are enough, we schedule the UE only in slot 1 & 3
so it has 6 slots until slot 7 to send PUCCH.
Before this commit, the DLSCH scheduler would construct the MAC PDU by
reading RLC data into a memory on the stack, and then construct the PDU
with CEs first. There are at least two problems:
- we need to keep track of the exact number of bytes of CEs (cumbersome)
to calculate the number of MAC SDUs to include
- we needlessly copy data around.
This commit does the following instead:
- write all CEs first (no need of keeping track of this in DLSCH and a
separate function)
- then read MAC SDUs directly into nFAPI as much as possible or
necessary, without recopying
According to SCF222, a single PDCCH allocation groups DCIs that are
within the same BWP and CORESET. Therefore, if we want to allocate
multiple DCIs, we need to decouple PDCCH allocation and DCI (previously
jointly done in nr_configure_pdcch()), especially to be forward
compatible.
***Note that as of this commit, we would still allocate different PDCCH
PDUs for multiple UEs (which we do not support yet, anyway)***
nr_configure_pdcch(): simply take out DCI allocation.
nr_generate_Msg2(): separately allocate dci_pdu in common RA SS, and
rename DCI payload variable. Also, reorganize the function so that it is
first checked for CCE allocation and messages nFAPI messages are
allocated afterwards.
nr_schedule_ue_spec(): separately allocate dci_pdu in UE-specific SS.
Rename DCI payload variable.
nr_schedule_ulsch(): separately allocate dci_pdu in UE-specific SS.
Rename DCI payload variable.
nr_fill_nfapi_dl_sib1_pdu(): separately allocate dci_pdu in common SS.
Initially, filling the PDCCH and PDSCH nFAPI messages was split into a
separate function (in an attempt to keep the code structure similar to
LTE). However, this proved as not helpful: the nr_fill_nfapi_dl_pdu()
just filled the messages, with a parameter list almost size as long as
the actual messages (because most parameters are kind of independent).
This made no sense, so we put it back.
Also, from an understanding POV, they just fill a message as specified
in SCF222, so it should not be a problem.
Robert Schmidt